diff mbox series

[3/8] scsi: don't memset the entire scsi_cmnd in scsi_init_command

Message ID 20220222140443.589882-4-hch@lst.de
State Superseded
Headers show
Series [1/8] bsg: don't include scsi_request.h in bsg-lib.h | expand

Commit Message

Christoph Hellwig Feb. 22, 2022, 2:04 p.m. UTC
Replace the big fat memset that requires saving and restoring various
fields with just initializing those fields that need initialization.

All the clearing to 0 ismoved to scsi_prepare_cmd as scsi_ioctl_reset
alreadly uses kzalloc to allocate a pre-zeroed command.

This is still conservative and can probably be optimizated further.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 drivers/scsi/scsi_lib.c | 61 ++++++++++++++++++++---------------------
 1 file changed, 29 insertions(+), 32 deletions(-)

Comments

John Garry Feb. 23, 2022, 12:21 p.m. UTC | #1
On 22/02/2022 14:04, Christoph Hellwig wrote:

Hi Christoph,

>   static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
> @@ -1586,10 +1558,35 @@ static blk_status_t scsi_prepare_cmd(struct request *req)
>   	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
>   	struct scsi_device *sdev = req->q->queuedata;
>   	struct Scsi_Host *shost = sdev->host;
> +	bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
>   	struct scatterlist *sg;
>   
>   	scsi_init_command(sdev, cmd);
>   
> +	cmd->eh_eflags = 0;
> +	cmd->allowed = 0;
> +	cmd->prot_type = 0;
> +	cmd->prot_flags = 0;
> +	cmd->submitter = 0;
> +	cmd->cmd_len = 0;
> +	cmd->cmnd = NULL;
> +	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
> +	cmd->underflow = 0;
> +	cmd->transfersize = 0;
> +	cmd->host_scribble = NULL;
> +	cmd->result = 0;
> +	cmd->extra_len = 0;
> +	cmd->state = 0;

I am just wondering did you consider using struct_group() for safety?

I don't think the scsi_cmnd members have special ordering apart from 
co-locating related members.

Here's how it could look (on top of yours):

---8<----

diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index 0b5c4606f641..51d2afa2d4c9 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -1549,18 +1549,8 @@ static blk_status_t scsi_prepare_cmd(struct 
request *req)

  	scsi_init_command(sdev, cmd);

-	cmd->eh_eflags = 0;
-	cmd->allowed = 0;
-	cmd->prot_type = 0;
-	cmd->prot_flags = 0;
-	cmd->submitter = 0;
-	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
-	cmd->underflow = 0;
-	cmd->transfersize = 0;
-	cmd->host_scribble = NULL;
-	cmd->result = 0;
-	cmd->extra_len = 0;
-	cmd->state = 0;
+	memset((char *)cmd + offsetof(struct scsi_cmnd, to_be_zeroed), 0, 
sizeof(cmd->to_be_zeroed));
+
  	if (in_flight)
  		__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);

diff --git a/include/scsi/scsi_cmnd.h b/include/scsi/scsi_cmnd.h
index 76c5eaeeb3b5..8d7a7186a3ab 100644
--- a/include/scsi/scsi_cmnd.h
+++ b/include/scsi/scsi_cmnd.h
@@ -73,10 +73,46 @@ struct scsi_cmnd {

  	struct rcu_head rcu;

-	int eh_eflags;		/* Used by error handlr */
-
  	int budget_token;

+	struct_group(to_be_zeroed,
+		int eh_eflags;		/* Used by error handlr */
+		int allowed;
+		unsigned char prot_type;
+		unsigned char prot_flags;
+		enum scsi_cmnd_submitter submitter;

...

---->8---

Thanks,
John
Christoph Hellwig Feb. 23, 2022, 12:39 p.m. UTC | #2
On Wed, Feb 23, 2022 at 12:21:10PM +0000, John Garry wrote:
> I am just wondering did you consider using struct_group() for safety?
>
> I don't think the scsi_cmnd members have special ordering apart from 
> co-locating related members.
>
> Here's how it could look (on top of yours):

Besides being ugly as hell I don't see the benefit.  Quite contrary,
a lot of these fields are properly initialized later and we can drop
the zeroing as well, but I'd rather do that separately and with a proper
audit.
John Garry Feb. 23, 2022, 12:56 p.m. UTC | #3
On 23/02/2022 12:39, Christoph Hellwig wrote:
> On Wed, Feb 23, 2022 at 12:21:10PM +0000, John Garry wrote:
>> I am just wondering did you consider using struct_group() for safety?
>>
>> I don't think the scsi_cmnd members have special ordering apart from
>> co-locating related members.
>>
>> Here's how it could look (on top of yours):
> 
> Besides being ugly as hell I don't see the benefit.  Quite contrary,
> a lot of these fields are properly initialized later and we can drop
> the zeroing as well, but I'd rather do that separately and with a proper
> audit.
> 
> .

ok, fine - so the init may be removed later.

For the record, advantages now could include:
- prob more efficient
- somewhat safer coding practice
- more condense C code

But, for sure, it's not so pretty to have this sub-struct.

Anyway, I don't feel too strongly about it and it was just a suggestion.

Thanks,
John
Christoph Hellwig Feb. 23, 2022, 12:58 p.m. UTC | #4
On Wed, Feb 23, 2022 at 12:56:32PM +0000, John Garry wrote:
> For the record, advantages now could include:
> - prob more efficient
> - somewhat safer coding practice
> - more condense C code
>
> But, for sure, it's not so pretty to have this sub-struct.
>
> Anyway, I don't feel too strongly about it and it was just a suggestion.

In general most of the zeroing here should go away entirely.  Right
now that code is a bit convolute due to the fake EH scsi command that
also uses scsi_init_command.  Once that is sorted out (I've just pinged
Hannes for his series) scsi_init_command can be folded into
scsi_prepare_cmd and a lot more of this can be cleaned up.
Bart Van Assche Feb. 23, 2022, 8:16 p.m. UTC | #5
On 2/23/22 04:58, Christoph Hellwig wrote:
> In general most of the zeroing here should go away entirely.  Right
> now that code is a bit convolute due to the fake EH scsi command that
> also uses scsi_init_command.  Once that is sorted out (I've just pinged
> Hannes for his series) scsi_init_command can be folded into
> scsi_prepare_cmd and a lot more of this can be cleaned up.

Several SCSI LLDs rely on the SCSI core zeroing the driver-private data.
I'd be more than happy if the code for zeroing of driver-private data 
would be pushed into the LLD drivers. This may require the introduction 
of a new per-command flag since that zeroing should only happen if 
scsi_queue_rq() decided to call scsi_prepare_cmd().

Bart.
Bart Van Assche Feb. 23, 2022, 11:03 p.m. UTC | #6
On 2/22/22 06:04, Christoph Hellwig wrote:
> Replace the big fat memset that requires saving and restoring various
> fields with just initializing those fields that need initialization.
> 
> All the clearing to 0 ismoved to scsi_prepare_cmd as scsi_ioctl_reset
                           ^
                    missing space?
> alreadly uses kzalloc to allocate a pre-zeroed command.
> 
> This is still conservative and can probably be optimizated further.
                                                  ^^^^^^^^^^^
                                                  optimized?
Anyway:

Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Christoph Hellwig Feb. 24, 2022, 4:27 p.m. UTC | #7
On Thu, Feb 24, 2022 at 08:28:48AM +0000, John Garry wrote:
>> +	if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
>> +		cmd->flags |= SCMD_INITIALIZED;
>
> Maybe I'm being dozy, but isn't this being cleared below *

with below you mean the

	cmd->flags &= SCMD_PRESERVED_FLAGS;

?  No, that doen't clear the flag, but all the others.
John Garry Feb. 24, 2022, 4:38 p.m. UTC | #8
On 24/02/2022 16:27, Christoph Hellwig wrote:
> On Thu, Feb 24, 2022 at 08:28:48AM +0000, John Garry wrote:
>>> +	if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
>>> +		cmd->flags |= SCMD_INITIALIZED;
>> Maybe I'm being dozy, but isn't this being cleared below *
> with below you mean the
> 
> 	cmd->flags &= SCMD_PRESERVED_FLAGS;
> 
> ?  No, that doen't clear the flag, but all the others.

Yeah, I was wrong as SCMD_INITIALIZED is included in (actually same as) 
SCMD_PRESERVED_FLAGS mask.

Sorry for the noise.

Cheers,
John
diff mbox series

Patch

diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c
index a1c18ba5e8d38..960795d469d8c 100644
--- a/drivers/scsi/scsi_lib.c
+++ b/drivers/scsi/scsi_lib.c
@@ -1163,45 +1163,17 @@  static void scsi_cleanup_rq(struct request *rq)
 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
 {
-	void *buf = cmd->sense_buffer;
-	void *prot = cmd->prot_sdb;
 	struct request *rq = scsi_cmd_to_rq(cmd);
-	unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
-	unsigned long jiffies_at_alloc;
-	int retries, to_clear;
-	bool in_flight;
-	int budget_token = cmd->budget_token;
-
-	if (!blk_rq_is_passthrough(rq) && !(flags & SCMD_INITIALIZED)) {
-		flags |= SCMD_INITIALIZED;
+
+	if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
+		cmd->flags |= SCMD_INITIALIZED;
 		scsi_initialize_rq(rq);
 	}
 
-	jiffies_at_alloc = cmd->jiffies_at_alloc;
-	retries = cmd->retries;
-	in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
-	/*
-	 * Zero out the cmd, except for the embedded scsi_request. Only clear
-	 * the driver-private command data if the LLD does not supply a
-	 * function to initialize that data.
-	 */
-	to_clear = sizeof(*cmd) - sizeof(cmd->req);
-	if (!dev->host->hostt->init_cmd_priv)
-		to_clear += dev->host->hostt->cmd_size;
-	memset((char *)cmd + sizeof(cmd->req), 0, to_clear);
-
 	cmd->device = dev;
-	cmd->sense_buffer = buf;
-	cmd->prot_sdb = prot;
-	cmd->flags = flags;
+	cmd->flags &= SCMD_PRESERVED_FLAGS;
 	INIT_LIST_HEAD(&cmd->eh_entry);
 	INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
-	cmd->jiffies_at_alloc = jiffies_at_alloc;
-	cmd->retries = retries;
-	if (in_flight)
-		__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
-	cmd->budget_token = budget_token;
-
 }
 
 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
@@ -1586,10 +1558,35 @@  static blk_status_t scsi_prepare_cmd(struct request *req)
 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
 	struct scsi_device *sdev = req->q->queuedata;
 	struct Scsi_Host *shost = sdev->host;
+	bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
 	struct scatterlist *sg;
 
 	scsi_init_command(sdev, cmd);
 
+	cmd->eh_eflags = 0;
+	cmd->allowed = 0;
+	cmd->prot_type = 0;
+	cmd->prot_flags = 0;
+	cmd->submitter = 0;
+	cmd->cmd_len = 0;
+	cmd->cmnd = NULL;
+	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
+	cmd->underflow = 0;
+	cmd->transfersize = 0;
+	cmd->host_scribble = NULL;
+	cmd->result = 0;
+	cmd->extra_len = 0;
+	cmd->state = 0;
+	if (in_flight)
+		__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
+
+	/*
+	 * Only clear the driver-private command data if the LLD does not supply
+	 * a function to initialize that data.
+	 */
+	if (!shost->hostt->init_cmd_priv)
+		memset(cmd + 1, 0, shost->hostt->cmd_size);
+
 	cmd->prot_op = SCSI_PROT_NORMAL;
 	if (blk_rq_bytes(req))
 		cmd->sc_data_direction = rq_dma_dir(req);